This paper (and an extensive supplementary report) considers how far cancer/risk factor associations based on epidemiology have been confirmed by evidence from 226 studies involving interventions other than smoking. Many are small, uncontrolled, of unrepresentative populations, concern cancer markers not cancer, and may involve combinations of agents. Many agents suspected of causing cancer are untested by intervention trials. For seven of 16 agents tested (fibre, folic acid, low-fat diet, riboflavin, zinc, vitamin Bs, and vitamin D), the evidence is clearly inadequate to confirm or deny the epidemiology, while the evidence relating to calcium only concerns biomarkers. For other agents, the evidence relating to cancer itself is weak. In studies where cancer is the endpoint, only three effects have been replicated: (a) selenium supplementation and decreased liver cancer incidence, (b) treatment by the retinoid etretinate and reduced bladder tumours in susceptible individuals, and (c) β-carotene supplementation and increased lung cancer incidence. Studies involving pre-cancerous conditions as the endpoint, which have a number of practical advantages, more frequently report benefits of intervention. Thus, oral pre-cancerous lesions can certainly be reduced by β-carotene, vitamin A, and other retinoids, and possibly by vitamin E. It also seems that retinoids can reduce pre-cancerous cervix, skin and lung lesions, that vitamin C and the NSAID sulindac can reduce colonic polyps, and that sunscreens can reduce solar keratoses. Our findings clearly show that the great majority of causal relationships suggested by epidemiology have not been validated by intervention trials. This may be partly due to lack of suitable studies of adequate size or duration, or to using single dietary compounds as agents that are by themselves not responsible for the epidemiologically-observed associations between diet and cancer. However, this lack of validation must cause concern in view of the markedly conflicting evidence on β-carotene and lung cancer between epidemiological and intervention studies. More intervention studies are needed, but in their absence, caution in interpreting epidemiological findings is warranted.